Independent Codes for Spatial and Episodic Memory in Hippocampal Neuronal Ensembles

See allHide authors and affiliations

Science  22 Jul 2005:
Vol. 309, Issue 5734, pp. 619-623
DOI: 10.1126/science.1114037


Hippocampal neurons were recorded under conditions in which the recording chamber was varied but its location remained unchanged versus conditions in which an identical chamber was encountered in different places. Two forms of neuronal pattern separation occurred. In the variable cue-constant place condition, the firing rates of active cells varied, often over more than an order of magnitude, whereas the location of firing remained constant. In the variable place-constant cue condition, both location and rates changed, so that population vectors for a given location in the chamber were statistically independent. These independent encoding schemes may enable simultaneous representation of spatial and episodic memory information.

Hippocampal neuronal ensemble activity appears to play an important role in the establishment of both spatial and nonspatial episodic memories, but there has long been controversy as to which of these parameters best characterizes the role of the hippocampal formation in mnemonic processes (1, 2). Although hippocampal neurons fire in a sparse (3), spatially selective (4, 5) manner, this activity can be weakly or strongly affected by other variables such as directional orientation (6, 7), specific sensory inputs (811), behavioral context (1217), and working memory (11, 1417). “Place fields” of a given neuron may change in magnitude, shift in position, or appear or disappear under different circumstances. All of these effects have been subsumed under the general terms “remapping” (1820) or “orthogonalization” (3, 21, 22), but it has not been determined whether these effects can occur independently across simultaneously recorded neurons. Here, we show that the hippocampal population vector can undergo two essentially independent forms of remapping: rate remapping, in which the locations of place fields remain unchanged but the firing rates of the cells change, and global remapping, in which the distributions of location and rate both take on statistically independent values. In vector terms, rate remapping implies that the population vectors span the same subspace and hence preserve information about location, whereas in global remapping the population vectors span statistically independent subspaces. In principle, rate remapping permits the generation of representations of unique episodes of experience while maintaining the integrity of the code for spatial location, whereas global remapping permits the distinction between similar experiences that occur in different spatial contexts.

Ensemble activity was recorded from 330 cells in CA3 and 487 cells in CA1 of the dorsal hippocampus in 10 freely moving rats by using arrays of 12 tetrodes, about half of which were located in CA3 and half in CA1 (23). Two experimental conditions were studied. In the variable cue-constant place condition (Fig. 1), the recordings were conducted in a single room, but the actual recording chamber was varied, either by changing the shape (circular versus square) or by changing the wall color (black versus white). In the variable place-constant cue condition (Fig. 2), geometrically identical (square) chambers of the same color (black) were used in two distinct rooms. All rats were familiarized with the experimental protocols and conditions before recording (24).

Fig. 1.

Constant place-variable cue condition. Color-coded rate maps for cells recorded when the cue configuration was changed by switching either between two colors (A and C) or between two shapes (B and D) at a constant location. (A and B) Recording sequence with the rate maps of the first cell in (C) and (D) as an example. Rate is coded on a color scale from blue (silent) to red (maximum rate). Pixels not sampled are white. The symbols above the maps indicate color and shape of the box and the cue card (cue card in contrasting color). (C and D) Rate maps for two complete sets of simultaneously recorded CA3 neurons [(C) rat 10708, day 3; (D) rat 10683, day 16]. Repeated trials with the same cues are now averaged. In (D), only the area common to both shapes is shown (indicated by the dotted square in the circle). Only the area common to both shapes is depicted. Each row shows data from one active neuron. Silent cells [six in (C) and six in (D)] are not shown. The left columns show data for the condition that gave maximal firing. For each cell, the rate scale corresponds to the peak firing rate in that condition (indicated to the left of the rate map). In the middle columns, the data from the opposite condition are plotted at the same firing-rate scale. The right columns contain the same data as the middle but are scaled to their own maximum values (indicated to the right of the rate maps). Symbols above columns indicate which color or shape gave the higher and lower rate (left and middle, respectively) for the cells underneath. Note that different experiences in the same place resulted in place fields similar in location and shape, but with strikingly different firing intensities, sufficient, in many cases, to make it appear that the field was absent when plotted at the same scale. See fig. S2 for rate maps of individual trials and fig. S4, A and B, for rate maps of simultaneously recorded CA1 cells.

Fig. 2.

Variable place-constant cue condition. Color-coded rate maps for tests in identical boxes in two different rooms (rat 10968, day 9). (A) Recording sequence and rate maps with the first cell in (B) as an example. (B) Rate maps for a complete set of simultaneously recorded CA3 place cells. The maps are averages for room A (A and A′) and room B (B and B′). Silent cells (n = 16) are not shown. Rows and columns are organized as in Fig. 1. Similar experiences in distinctly different places resulted in place fields with essentially zero correlation in location or rate. See fig. S3 for rate maps of individual trials and fig. S4C for rate maps of simultaneously recorded CA1 cells.

For each cell whose mean firing rate in at least one environmental condition was above an estimate of background noise (25), we compared the location and rate of firing in the different experimental configurations. Spatial similarity was expressed by computing the correlations between pairs of rate maps and comparing their centers of mass. These measures are independent of the relative magnitude of firing. Rate similarity was expressed by dividing the relative change in mean rate by the sum of the mean rates. Lastly, representational similarity in the population as a whole was estimated by stacking all rate maps (including those of the silent cells) into a three-dimensional matrix with the two spatial dimensions on the x and y axes and cell identity on the z axis (fig. S1C). The distribution of mean rates along the z axis for a given x-y location represents the population vector for that location. Comparing the entire set of population vectors between two trials provides an estimate of how much the ensemble code changed.

For both CA3 and CA1, the distributions of spatial correlations in the two variable cue-constant place conditions (i.e., black square versus white square and circle versus square) were so similar to the control conditions (first versus second exposure in the same box) that, without closer analysis, one might have concluded that there had been no substantive effect of cue-condition or environmental shape on the hippocampal code (Figs. 1 and 3A and Table 1). In CA3, the median spatial correlations in the black-white and square-circle conditions were 0.82 and 0.83, respectively. In CA1, spatial correlations were similar to those obtained in CA3 (black-white: 0.78, Z = 1.54, not significant) or slightly lower (square-circle: 0.65, Z = 3.92, P < 0.001). In both areas, there were only minor shifts in center of mass of the place fields (Table 1).

Fig. 3.

Spatial firing rate statistics for ensembles of hippocampal CA3 (A) and CA1 (B) neurons comparing the following conditions: two visits to the same box in a constant location, visits to black and white boxes in a constant location, visits to circular and square boxes in a constant location, and visits to identical boxes located in two distinctly different places. For each pair of conditions, columns from left to right show frequency distributions for correlations in the spatial firing patterns (PF correlation), shifts of the place-field centers (Δ center of mass), changes in the mean firing rates (Δ rate), and correlations between spatial population vectors (PV correlation). Rate change is expressed as the unsigned difference between the rates in the two conditions, divided by their sum. On this asymptotic scale, 0.5 corresponds to a threefold difference, whereas 0.8 represents a ninefold difference. Whereas a variable place exerted a marked influence on both rate and firing location, changes in sensory cues in a fixed place resulted only in substantial changes in the firing rates, especially in CA3, with little change in the relative spatial firing distribution, implying that cues and locations are coded independently. See fig. S5 for additional comparisons in the constant cue-constant place configuration.

Table 1.

Spatial firing rate statistics.

Condition CA3 CA1
Median place field correlation
Constant cue and place 0.94 0.88
Color 0.93 0.87
Shape 0.97 0.93
Room 0.92 0.84
Black-white 0.82 0.78
Circle-square 0.83 0.65
Room1-room2 –0.10 0.00
Median center of mass change (cm)
Constant cue and place 3.3 5.6
Color 4.3 5.9
Shape 2.3 4.1
Room 3.6 6.9
Black-white 9.8 8.6
Circle-square 7.4 10.2
Room1-room2 34.5 26.8
Median relative firing rate change (unsigned rate difference divided by rate sum)
Constant cue and place 0.12 (1.3-fold) 0.14 (1.3-fold)
Color 0.10 (1.2-fold) 0.13 (1.3-fold)
Shape 0.12 (1.3-fold) 0.13 (1.3-fold)
Room 0.14 (1.3-fold) 0.16 (1.4-fold)
Black-white 0.59 (3.9-fold) 0.19 (1.5-fold)
Circle-square 0.74 (6.9-fold) 0.30 (1.9-fold)
Room1-room2 0.97 (61.6-fold) 0.42 (2.4-fold)
Median population vector correlation
Constant cue and place 0.91 0.83
Color 0.91 0.85
Shape 0.94 0.85
Room 0.91 0.75
Black-white 0.58 0.73
Circle-square 0.24 0.57
Room1-room2 –0.03 0.25

In contrast, although the location of firing remained almost constant, the distributions of mean firing rates changed, especially in CA3. In this subfield, a substantial proportion of cells showed pronounced rate differences between conditions, with little or no change in the shape or location of the place field (Figs. 1 and 3A and Table 1). The median rate change between the black and white chambers was 3.9-fold; between square and circle, it was 6.9-fold. The median rate change was significantly higher than that for repeated tests with the same cues (difference/sum ratios; black-white, Z = 8.58 and P < 0.001; square-circle, Z = 7.54 and P < 0.001). The change in individual firing rates was also reflected as a pronounced shift toward lower values in the population vector correlations (Fig. 3A and Table 1) (black-white: Z = 22.48, P < 0.001; square-circle: Z = 19.40, P < 0.001). In CA1, the changes in relative firing rate were less pronounced (CA3 versus CA1: Z was 6.39 and 5.49 for black-white and square-circle, P < 0.001) and the population vectors were less different (Z was 11.14 and 13.72, respectively, P < 0.001) (Fig. 3B and Table 1).

In the variable place-constant cue condition, the distributions of spatial and population vector correlations were both shifted to values near zero, and the overall firing rate differences were large (Figs. 2 and 3 and Table 1). In CA3, firing locations in the two boxes were totally uncorrelated. Place field correlations were also skewed toward zero in CA1, albeit with a significant tail toward higher values (Fig. 3A) (Kolmogorov-Smirnov test, P < 0.05) (Fig. 3B). The change in firing locations was accompanied by pronounced rate changes in the same cells. In CA3, the majority of neurons showed more than 10-fold rate changes, and the population vector correlations for different locations were clustered around zero. In CA1, the extent of rate change varied widely within the population, and population vector correlations ranged between 0 and 0.5. In each animal, the population vectors were more dissimilar in CA3 than CA1.

To summarize, when the same prominent environmental features were encountered in two distinct places, the CA3 population vectors for locations (relative to the reference frame of the constant feature) were uncorrelated. This is consistent with previous studies reporting orthogonalization of hippocampal population activity, especially in CA3, after dislocation of the local reference frame (22, 2628). When the rats experienced changes in prominent features in the same spatial context, however, the hippocampus encoded these changes primarily by changing the values of the components of the spatial population vectors, without much, if any, change in the distribution of the firing locations (29). The latter finding raises the possibility that much of the previous evidence for single-location remapping is rate-based. Rate remapping may be the cause of direction-specific (6, 30) and trajectory-specific (1416) firing as well as task-induced changes in population activity (12, 13, 31). However, certain training regimes may provoke global remapping even in a single spatial environment (1820, 32). Whether the environments are then encoded as distinct spatial reference frames remains to be determined.

The existence of independent population codes for location and cue configurations implies that hippocampal cell ensembles may simultaneously convey information related to where an animal is located and what is currently present in that location. Presynaptic dynamics (33) as well as nonlinear postsynaptic responses may render some neurons sensitive to which inputs are active but not to presynaptic firing rate per se. Cells with such properties might be able to transmit information about the location of the animal irrespective of current sensory input. In contrast, cells that are sensitive to the rate function of afferent neurons might discriminate different events occurring at a given location. The preservation of both types of information in the hippocampal output may form the basis of its key role in episodic memory (2, 34).

Supporting Online Material

Materials and Methods

Figs. S1 to S5

Tables S1 and S2

References and Notes

View Abstract

Stay Connected to Science

Navigate This Article